Capturing Temporal Dynamics of Implicit Feedbacks for Collaborative Filtering by Using Deep Recurrent Neural Networks

Abstract

Collaborative Filtering (CF) is one of successful methods for generating recommendations. However, conventional CF method is not good at capturing sequential behaviors of users. Though it is obvious that users preference must be affected by experiences of the users in the past. For example, if a user watched a movie and then liked the movie significantly, the user will be interested in movies whose actor/actress/director is the same. In other words, each users preference towards items is evolving with time dynamically and these temporal dynamics should be noticed. In this paper, we treat users rating histories as sequential purchasing behavior and utilize such sequential behavior for recommendation task. We use Deep Recurrent Neural Networks (DRNN) to model those purchasing sequences. Experiments on the MovieLens dataset show improvements over previous reported results and demonstrate that our method can utilize users purchasing behavior data for collaborative filtering while capturing the evolving behaviors and tastes of users better by modeling temporal dynamics implicitly.